Study On Full-Duplex Cognitive Anti-Jamming Based On Reinforcement Learning

Due to the rapid development of modern information warfare,spectrum resources are increasingly scarce.At the same time,because the jamming equipment becomes more and more serious intelligent,the effect of the traditional anti-jamming technology is getting weaker.Cognitive radio(CR)can solve the scarcity of spectrum resources and combine it with full-duplex technology to form a full-duplex cognitive anti-jamming(SCAJ)technology,which can improve the anti-jamming ability in complex electromagnetic environment.It can transmit data while sensing surrounding jamming.It can improve communication anti-jamming ability.Therefore this thesis analyzes the new method of jamming sensing based on multi-slot.This thesis focuses on full-duplex cognitive anti-jamming technology,and the effects of the multi-slot jamming sensing.The performance of the reinforcement learning(RL)cognitive anti-jamming method under the actual jamming sensing and the cognitive anti-jamming effects based on deep reinforcement learning(DRL)is studied.Firstly,in order to solve the performance of single-slot jamming sensing is not accuracy,a multi-slot jamming sensing method using improved energy detection is proposed.In this method,multi-slot based on the improved energy detector is used to jamming sensing.The sensing probability and the false alarm probability are derived.Based on the multi-slot sensing results of local radios,the cooperative spectrum sensing performance of the full-duplex network is studied,and the total error rate expression and the system average capacity expression under the "Majority" fusion criterion are obtained.Secondly,most of the research is studied anti-jamming under ideal sensing.Therefore,this thesis studies the Stateless Q learning in full-duplex cognitive anti-jamming of actual jamming sensing results.The interaction between a single station and a jammer is modeled as Stateless Q learning,and the false alarm probability and false detection probability of the actual jamming sensing are introduced into the reward function of Stateless Q learning,thereby solving the problem of channel and power selection.Further,a single station considers the influence of other stations to improve the throughput and fairness of the network,and studies the multi-agent Q learning(MAQL)algorithm based on full-duplex cognitive anti-jamming for shared reward.Finally,if the radio cannot fully observe the dynamic jamming environment,it will reduce the anti-jamming ability.Aiming to solve this problem,this thesis proposes a full-duplex cognitive anti-jamming based on Double DQN method,combined Double DQN with full-duplex cognitive anti-jamming technology.In this method,the system throughput obtained based on the false alarm probability and the detection probability of the actual jamming sensing is taken as the historical information,and the radio learns the better anti-jamming strategy through the Double DQN aim to improve anti-jamming performance.